Cloudflare Agents deepens long-running support

This is Cloudflare leaning into the boring plumbing that makes agents usable in production: state, recovery, and connection continuity. It is less flashy than a model launch, but much more important if you want agents that survive crashes, redeploys, and user re-entry. Durable execution and checkpointing reduce fragility for long runs, WebSocket support and reconnectable sessions keep live work intact, and persistent sessions plus sub-agents suggest Cloudflare wants to own more of the orchestration layer.